Recently, Apple’s iPhone 7 has officially gone on sale, and the iFixit team in San Luis Obispo, California, has “disassembled” the iPhone 7 Plus, and their teardown images and commentary report quickly took over the tech media worldwide.

The main board of the iPhone 7 Plus integrates a very small FPGA, measuring 2mmx2mm. This chip is the low-density, low-power small package FPGA called iCE40, which was launched by Lattice a few years ago, model number iCE5LP4K, with a density of 3.5K LUT-4 equivalent logic resources. Surrounding components include audio CODEC, audio amplifiers, diversity receivers, along with a barometric pressure sensor on its back, NFC controller, WIFI/Bluetooth controllers, etc.

According to some technical experts’ speculations, the function realized by this FPGA may be the function of a Sensor HUB, which is to aggregate and preprocess various sensors (possibly including bus control, bus conversion, time division multiplexing, interface conversion, etc.), and interact with the main processor CPU for further calculations when necessary, reducing the monitoring and calculation frequency of the main processor CPU related modules, thereby lowering power consumption and improving processing performance.

Of course, some other technical experts have different opinions. They believe that the main role of this FPGA is for controlling USB Type C. Currently, due to standards and other reasons, no ASSP manufacturers have provided complete solutions, while Lattice takes advantage of the characteristics of FPGA to become one of the first suppliers of USB Type C. This is also the advantage of FPGA compared to ASSP, which accelerates the time to market and helps customers save time costs. However, until Apple or Lattice publicly reveals the design intention, these are just speculations.

The list of disassembled components of the iPhone 7 Plus is as follows:

Apple/Cirrus Logic 338S00105 Audio Codec

Cirrus Logic 338S00220 Audio Amplifier(x2)

Lattice Semiconductor ICE5LP4K FPGA

Skyworks 13702-20 Diversity Receive Module

Skyworks 13703-21 Diversity Receive Module

Avago LFI630 183439

NXP 610A38

Apple A10 Fusion APL1W24 SoC

Samsung 3 GB LPDDR4 RAM

Qualcomm MDM9645M LTE Cat. 12 Modem

Skyworks 78100-20

Avago AFEM-8065 Power Amplifier Module

Avago AFEM-8055 Power Amplifier Module

Universal Scientific Industrial O1 X4

Bosch Sensortec BMP280 Barometric Pressure Sensor

Toshiba THGBX6T0T8LLFXF 128 GB NAND Flash

Murata 339S00199 Wi-Fi/Bluetooth Module

NXP 67V04 NFC Controller

Dialog 338S00225 Power Management IC

Qualcomm PMD9645 Power Management IC

Qualcomm WTR4905 Multimode LTE Transceiver

Qualcomm WTR3925 RF Transceiver

TDK EPCOS D5315

Texas Instruments 64W0Y5P

Texas Instruments 65730A0P Power Management IC

This time, Lattice’s FPGA devices have never entered Apple’s supply chain before, this is a groundbreaking first, and it is also the first time Apple has introduced FPGA devices in the iPhone. Why? According to common sense, with Apple’s annual sales of 100 million iPhones, if they want to minimize costs, they should use ASIC chips as much as possible; traditionally, FPGA has usually only been used in low-volume markets, so Apple’s design this time has indeed attracted everyone’s interest.

Years ago, Samsung’s Galaxy S4 and S5 also took the lead in using Lattice’s same FPGA devices, and at that time the main application direction was for infrared recognition and QR code recognition. However, in the S5 and later Galaxy models, Samsung replaced FPGA with other solutions, so it did not cause a stir in the industry, but the significance of Apple using FPGA is greatly different.

Apple’s iPhone 7 adopting the FPGA solution has set a benchmark in the smartphone industry, which also means that the FPGA device, which is usually not prominent, has appeared so conspicuously in the consumer electronics market, winning widespread attention. This situation was completely impossible many years ago.

At least we can draw a simple conclusion: in the future, whether or not to integrate FPGA may become a watershed that distinguishes high-end smartphones; perhaps in the future, we will find more consumer electronics will apply FPGA. This should be good news for the FPGA industry.

The Development History of Lattice iCE Series

The chip used by Apple this time belongs to the ICE series. In 2012, Lattice released the first FPGA for mobile applications, iCE40. As for the naming of iCE products, it is consistent with the product’s characteristics, which is low power like ice. Before Lattice launched iCE, the consumer market accounted for less than 5% of its total revenue, while now consumer sales are close to 30% of sales, all thanks to the new market applications.

With the rapid evolution of handheld market technologies, many functions are still in the experimental stage without corresponding protocols and stable markets, and ASSP manufacturers are unwilling to develop them. Therefore, in the window period of half a year to a year, FPGA can help customers solve transitional issues. Lattice compares iCE series to a “universal solution,” meaning it is the most flexible wearable device. In addition, FPGA can help customers save debugging time because once ASSP has a bug, customers can only wait for the next batch of products, while FPGA can be modified at any time, shortening the development cycle.

In 2015, Lattice launched the iCE40 UltraLite for mid-range phones and wearable devices, as the name suggests, it is a simplified version of iCE40 Ultra, but the configuration has not been reduced; the biggest change is that the storage space is reduced, and the logic units are reduced from 1k/2k/3k to 0.64k and 1.2k, allowing customers to choose according to their applications. For example, for wristbands or mid-range phones, UltraLite can achieve low power consumption, differentiation, and low-cost product design. The package size is as small as 1.4×1.4×0.45mm, which is reduced by 60% compared to competitors’ products; power consumption is only 35uA, which is 30% lower than that of competitors; meanwhile, it has the highest hardware integration level.

The application range targeted by the iCE series includes: wearable devices, pedometers, always-on navigation, voice input, LED breathing effects, schedule reminders, activity monitoring, infrared control, user recognition, etc. Meanwhile, UltraLite also supports various industrial applications, including handheld tools, POS, and other traditional industrial applications, as well as portable blood pressure monitors and other handheld medical devices, dental, gastroscope, and other internal medical applications.

The FPGA Market is Entering a Rapid Growth Period

FPGA is known as the “universal chip”; from a technical perspective, it is a programmable product that can be reused without changing the hardware composition of the chip itself. Depending on market changes, it can realize different functions on the same carrier. It sounds great, but it’s difficult to develop; manufacturers that can produce FPGA devices must be top-notch, as semiconductor giants like Intel and Texas Instruments have also attempted to enter this market but have failed for various reasons. Currently, globally, only a handful of companies can mass-produce FPGA. Xilinx, Altera, Lattice, and Microsemi are all well-known American companies.

FPGA is a semi-custom circuit in the field of application-specific integrated circuits (ASIC), with advantages including quick production, modifiability, the ability to correct errors in programs, and lower costs. Engineers can reconfigure the internal logic modules and I/O modules of FPGA to achieve the desired logic. FPGA also has static reprogrammability and dynamic reconfiguration characteristics, allowing hardware functions to be modified through programming like software. FPGA can perform the functions of any digital device, even high-performance CPU can be implemented using FPGA. FPGA is like a blank sheet of paper: when powered on, the FPGA chip reads data from the EPROM into the internal programming RAM, and once configured, it enters working status; when power is off, FPGA returns to a blank state, and the internal logic relationships disappear, so FPGA can be reused. Using FPGA for digital circuit development can greatly shorten design time, reduce the area of PCB printed circuit boards, and improve system reliability. FPGA can embed hard-core CPU or soft-core CPU, achieving both digital logic and adapting to embedded development.

In 2014, the global FPGA market reached a total size of 5 billion USD, with China’s market share nearing 1.7 billion USD, accounting for one-third of the global market. Analysts predict that from 2015 to 2020, the global FPGA market’s compound annual growth rate will be 9%, and by 2020, the global FPGA market size will reach 8.4 billion USD. Currently, FPGA is in a rapidly growing market state, with growth far exceeding other chip markets; at the same time, the average gross margin of the FPGA industry is considerable, with market data analysis indicating that the average gross margin of the industry is greater than 60%.

The FPGA industry also needs a larger market scale to attract more users. It is expected that as FPGA production gradually increases, costs will further decrease, and its market share will continue to grow.

Currently, the main application fields of FPGA are communications, industrial control, national defense, and consumer electronics. In recent years, one of the most noticeable trends in FPGA is the continuous expansion of application fields.

In the traditional application market of FPGA, communication is gradually achieving high-speed and complex protocols, while consumer electronics applications focus on low power consumption and low cost. In addition, FPGA is also widely used in medical electronics, security, video, industrial automation, voice networks, semiconductor manufacturing equipment, and home appliances. FPGA applications in industry mainly involve four areas: motor control, industrial networks, machine vision, and robotic control. Taking motor control as an example, the focus is on reducing noise, minimizing vibration, lowering EMI, while achieving higher control accuracy and reducing energy consumption.

Currently in the global market, Xilinx and Altera dominate FPGA technology and market. The two companies hold nearly 90% of the market share, with over 6000 patents; the remaining market share is mainly occupied by Lattice and Microsemi, which also have over 3000 patents. In 2014, Xilinx and Altera generated revenues of 2.38 billion USD and 1.93 billion USD, respectively, while Lattice and Microsemi (only FPGA portion) generated revenues of 366 million USD and 275 million USD, respectively. These four leading FPGA market players almost monopolize the FPGA market share.

The FPGA market outlook is attractive, but the barriers to entry are high, unmatched in the chip industry. More than 60 companies have invested billions of dollars in attempts to reach the FPGA summit, including giants like Intel, IBM, Texas Instruments, Motorola, Philips, Toshiba, and Samsung, but only four companies in Silicon Valley, Xilinx, Altera, Lattice, and Microsemi, have succeeded. The aforementioned four companies hold over 9000 patents, creating a formidable technological barrier.

Moore’s Law and Intel’s Acquisition of Altera for Cloud Computing Ambitions

Last year, Intel acquired Altera, the second-largest FPGA company, for 16.7 billion USD. At that time, Intel’s CEO Brian Krzanich stated that Intel’s growth strategy is to expand core assets to enter high-profit, complementary market spaces. Through this acquisition, Intel can push Moore’s Law toward better and stronger next-generation solutions. Enterprise users expect to achieve cost-effective performance in enhanced networks, large cloud data centers, and the Internet of Things, which is precisely what Moore’s Law and Intel’s joint innovation with Altera can achieve.

So why is Intel willing to pay such a high price (almost 8 times the sales rate) to acquire this company? The author notes that in 2014, Microsoft collaborated with Amazon, Google, and several top universities to publish a long paper titled “Enhancing Configurable Architectures for Large-Scale Datacenter Services” in IEEE. The participants also included technical support from Altera and Quanta, and the project mentioned in the paper was named “Catapult,” which is an aggressive project actively promoted by Microsoft Research. In the era of cloud computing, Microsoft Research has been actively considering how to improve the current cloud infrastructure. Due to the rapid growth of cloud infrastructure, once cloud data centers exceed a certain scale, they will face challenges of stability and sustainability. Microsoft Research is actively exploring whether other alternative technologies can achieve sustainable growth of cloud data centers, and this technology is FPGA field programmable gate arrays.

Since the first FPGA product was born in 1985, FPGA applications have gradually expanded from digital signal processing and high-speed serial transceivers to embedded processors. FPGA has been widely used in communications, consumer electronics, industrial control, testing, and measurement in the microelectronics and electrical fields, but has hardly been applied in large data centers.

Microsoft Research announced in June 2014 that it successfully completed tests using FPGA technology (rather than GPU technology) in the Bing cloud data center. The results include being 40 times faster than traditional servers in processing Bing’s custom algorithms, 2 times faster than Bing’s existing systems, and being able to reduce the number of existing servers by half. FPGA technology accelerates the processing of large amounts of data on servers, helping to solve the big data problem and meet the enormous demands of distributed workloads. The paper “Enhancing Configurable Architectures for Large-Scale Datacenter Services” discusses how to test FPGA technology on Bing.

In September, Baido quickly followed suit, announcing that it would use FPGA technology in its data centers.

This means that the future data center market space may rely on FPGA to achieve, and for Intel, which previously had a low market share in the communication and network markets, the combination with Altera is expected to open up this market space.

The application of FPGA in cloud data centers will evolve from discrete use of CPU and FPGA to packaged use of CPU and FPGA, and then to integrated use of CPU and FPGA. It will play an important role in convolutional neural network algorithms for image recognition, encryption algorithms for security control, and compression algorithms for big data processing. Microsoft’s research indicates that FPGA consumes about 1/10 of the power of high-end GPU, but the image processing capability of high-end GPU is 2 to 3 times that of FPGA, while clusters composed of multiple FPGA can achieve the image processing capability of GPU while maintaining low power consumption characteristics.

According to Intel’s estimates, by 2020, 1/3 of cloud data center nodes will adopt FPGA technology.

In terms of the Internet of Things, Altera’s FPGA technology combined with Intel’s Atom chip is expected to replace existing technologies in the industrial automation control market and advanced driver assistance system market, generating 11 billion USD in incremental serviceable market by 2020. By adopting FPGA technology, users can programmatically replace existing application-specific standard circuits (ASSP) and application-specific integrated circuits (ASIC), adding new functions, improving performance, reducing costs, and shortening time-to-market by 50%.

Currently, the competition in the FPGA market is undergoing changes, and the application fields are continuously expanding. Previously, FPGA was largely dominated by the communication market, but the development of industrial intelligence, automotive electronics, and the Internet of Things has led to a significant increase in the demand for flexible programming FPGA, and FPGA manufacturers tend to shift towards solution providers, actively developing consumer markets. Collaborating with large companies like Intel may determine the market’s leading solution position. Once Intel and Altera’s solutions dominate the market, programmers and engineers will flock to this technological direction, further consolidating Intel’s market position.

In June 2014, Intel announced the integration of industry-leading Xeon processors and FPGA into a single structure, compatible with standard Xeon processor slots. Intel stated that FPGA will provide its customers with stronger algorithm execution capabilities and higher performance. Industry research has shown that FPGA-based accelerators can bring more than 10 times performance improvement, while Intel claims it can provide an additional two times improvement on top of that, using a super-low latency interface between FPGA and processors.

This means that Intel can bring more than 20 times speed boost to server-based applications by adopting FPGA!

By September, Altera began collaborating with Baido to use FPGA and convolutional neural networks (CNN) in Baido’s deep learning applications. Baido stated that FPGA acceleration has enormous application potential and will help FPGA penetrate into mainstream heterogeneous computing fields. Utilizing Altera FPGA hard cores like IEEE 754 floating-point multipliers and adders, servers and data centers can meet increasingly complex demands in search, big data, and deep learning.

It is estimated that the power consumption of global data centers accounts for 1%-10% of total global power consumption, and with the growth of data volume, power consumption is also rapidly rising. Now, ensuring faster computing power while reducing energy consumption growth has become a significant economic and ecological issue. Today’s data centers are built on cheap power, and the size and throughput of data centers are limited by available power input and heat dissipation capacity. Companies like Microsoft, Google, Facebook, and eBay, which rely heavily on servers, have sufficient motivation to strive for improved code execution speed and reduced electricity costs.

Meanwhile, as Moore’s Law approaches its end after nearly 50 years, the power issues of our single-core processors have reached their limits, and the rate of power consumption increase has exceeded the rate of performance improvement. Then we have dual-core, quad-core, and even octa-core chips. Finally, to compensate for the lack of continuous improvement in semiconductor processes, we are looking for other methods. Simply waiting for better semiconductor materials to solve the problems of data centers is not a viable option.

Why can FPGA significantly improve computing power with lower power consumption? For specific algorithms, hardware based on FPGA can provide high granularity parallel computing—lower latency, higher throughput, and lower power consumption.

It is also worth noting that Intel is not the first company to plan mass production of heterogeneous processors with FPGA. Xilinx has been fighting in this market for several years with its Zynq series, which uses Xilinx FPGA-based ARM processors. Although Zynq is not a data center-level processor, the gap is still significant, but Xilinx can leverage the experience and tool accumulation gained from Zynq to secure a place in the future low-power server market.

From the perspective of the five major markets of FPGA, industries like communications have occupied about 70% of the market share, while computer applications and medical devices are developing rapidly, and the low-cost, low-power FPGA is entering the consumer market. With the development of communications, industrial control, robotics, video analysis, and even consumer markets, the market size of FPGA is even larger. Recently, there have been views expressing that FPGA may replace CPU to become the main chip, and whether FPGA, with its high investment threshold, can continue to develop steadily in domestic markets remains a question.

Can FPGA become the antidote to the failure of Moore’s Law?

Is Lattice the Only Supplier of Consumer Electronics FPGA?

Previously, the author mentioned that the entire global FPGA market is currently completely monopolized by four American companies, and Altera and Xilinx monopolize around 88% of the market. For domestic Chinese companies, FPGA is almost a blank space, and they will still rely heavily on imports now and in the future. If another crisis like the ZTE ban occurs, it could be disastrous for any company. The rapid growth of FPGA will pose more challenges for China. When can China fill this important gap in the semiconductor puzzle remains a focus for both the government and major manufacturers.

However, according to product information in the author’s possession, some domestic companies have successfully developed compatible devices with Lattice iCE series and are already selling them in the market. Therefore, the author is not too worried that Chinese people will eventually conquer the FPGA fortress built by Americans. This company not only has compatible Lattice devices but also more compatible devices with Altera mid-range chips. (To avoid advertising suspicions, the company name is omitted).

Samsung’s Galaxy series has also featured this domestic manufacturer, which can indirectly confirm that it is not reverse engineering; otherwise, such a large manufacturer like Samsung would not procure it due to concerns over intellectual property issues.

The Future of FPGA

As the demand for intelligent markets changes rapidly, the trend of reduced custom chip (SoC ASIC) projects has become irreversible. The investment in ASIC solutions has significantly increased, with longer cycles and greater market risks. On the other hand, FPGA technology is gradually moving towards the mainstream.

FPGA has advantages in parallel computing and can replace some ASIC and DSP in high-performance, multi-channel fields. The parallel computing of FPGA can optimize traditional ASIC solutions in multi-channel processing.

With a critical point of 50,000 chips, FPGA has a trend to replace ASIC in high-value, relatively small-volume, multi-channel computing devices (such as radars, spacecraft, routers), which will be a territory Xilinx firmly holds.

FPGA development cycles are 55% shorter than ASIC, allowing for rapid market capture. This may be the most important trump card for Lattice in the large-volume consumer electronics market.

As for the ultimate goal of Altera FPGA under Intel, it should be to leverage its advantages in high-performance, low-power, and deep learning to ultimately integrate into CPU technology, achieving the future of heterogeneous computing data centers.

From the perspective of capital markets, previously, Ziguang Group just completed a minor stake in Lattice. Currently, Lattice (LSCC) is listed on NASDAQ, with a total market value of less than 800 million USD, compared to Altera‘s 16.7 billion USD acquisition price; this size could be an ideal acquisition target. There are only these four markers in the global FPGA industry (Xilinx, Altera, Lattice, and Microsemi). Intel has already acquired Altera, and Xilinx, as the industry leader, is unlikely to sell for the time being. Selling one of the remaining two scarce markers means one less. Riding the trend of entering the iPhone 7 supply chain, it is uncertain whether Lattice will see its market value soar this year and become a target for big players. However, it seems that the U.S. government will most likely not approve the acquisition of such a key technology-sensitive company by a Chinese company, so domestic FPGA companies must accelerate their pace to strive for early replacement of imports, so as to break free from being constrained.

Semiconductor Industry Observation（icbank）

Semiconductor Industry Observation: The first vertical media in the semiconductor industry 【www.icbank.cc】

Semiconductor Circle（icquan）

The semiconductor circle is a network of leading entrepreneurs, entrepreneurs, investors, and ICT technology talents in the semiconductor industry.

Semiconductor Industry Alliance（icunion）

Collaboration Methods

1. Collaborative Promotion (WeChat: 97915500)

2. Paid Contributions/Tips (Email: [email protected])